27 Oct
2018
27 Oct
'18
5:46 a.m.
On Fri, Oct 26, 2018 at 05:06:48PM -0600, Warner Losh wrote:
On Fri, Oct 26, 2018 at 4:57 PM Angelo Papenhoff
<aap(a)papnet.eu> wrote:
Yeah, it's nice when Ken speaks up. I'm always nervous when I post
because I know Ken wants 100% signal.
It would be nice if we could get Rob Pike here, he's the right combination
of brilliant and kinda snotty, he doesn't suffer fools at all. I loved
Rob's point that if you think that you need threads your processes are
too fat. He's mostly right, in theory, but in practice you can't share
memory via mmap across processes at the same low cost as threads. If you
have N processes sharing the same memory you have N*pages page table
entries; if you have threads you have 1*pages page table entries (for
the shared memory part). Back when I was active in kernel development
I tried to get Linus to do a page table scheme that would allow more
than one page table per process so you could have private tables and
shared tables. Dunno if he did that.
But it would be nice to have Rob here, he's sort of old school but sort
of the next gen of Bell Labs. And a smart cookie.
And to Angelo, Warner is right, I'd keep that email.
--lm
On 26/10/18, Ken Thompson wrote:
Today, sir, you have won this part of the internet. :) nice job.
Thanks, You've always been an inspiration to me!